High-throughput B-cell sequencing has opened up new avenues for investigating complex mechanisms underlying our adaptive immune response. These technological advances drive data generation and the need to mine and analyze the information contained in these large datasets, in particular the identification of therapeutic antibodies (Abs) or those associated with disease exposure and protection. Here, we describe our efforts to use artificial intelligence (AI)-based image-analyses for prospective classification of Abs based solely on sequence information. We hypothesized that Abs recognizing the same part of an antigen share a limited set of features at the binding interface, and that the binding site regions of these Abs share share common structure and physicochemical property patterns that can serve as a "fingerprint" to recognize uncharacterized Abs. We combined large-scale sequence-based protein-structure predictions to generate ensembles of 3-D Ab models, reduced the Ab binding interface to a 2-D image (fingerprint), used pre-trained convolutional neural networks to extract features, and trained deep neural networks (DNNs) to classify Abs. We evaluated this approach using Ab sequences derived from human HIV and Ebola viral infections to differentiate between two Abs, Abs belonging to specific B-cell family lineages, and Abs with different epitope preferences. In addition, we explored a different type of DNN method to detect one class of Abs from a larger pool of Abs. Testing on Ab sets that had been kept aside during model training, we achieved average prediction accuracies ranging from 71-96% depending on the complexity of the classification task. The high level of accuracies reached during these classification tests suggests that the DNN models were able to learn a series of structural patterns shared by Abs belonging to the same class. The developed methodology provides a means to apply AI-based image recognition techniques to analyze high-throughput B-cell sequencing datasets (repertoires) for Ab classification.

Quantitatively predicting changes in drug sensitivity associated with residue mutations is a major challenge in structural biology. By expanding the limits of free energy calculations, we successfully identified mutations in influenza neuraminidase (NA) that confer drug resistance to two antiviral drugs, zanamivir and oseltamivir. We augmented molecular dynamics (MD) with Hamiltonian Replica Exchange and calculated binding free energy changes for H274Y, N294S, and Y252H mutants. Based on experimental data, our calculations achieved high accuracy and precision compared with results from established computational methods. Analysis of 15 micros of aggregated MD trajectories provided insights into the molecular mechanisms underlying drug resistance that are at odds with current interpretations of the crystallographic data. Contrary to the notion that resistance is caused by mutant-induced changes in hydrophobicity of the binding pocket, our simulations showed that drug resistance mutations in NA led to subtle rearrangements in the protein structure and its dynamics that together alter the active-site electrostatic environment and modulate inhibitor binding. Importantly, different mutations confer resistance through different conformational changes, suggesting that a generalized mechanism for NA drug resistance is unlikely.

United-residue models of polypeptide chains [3, 5, 19-22, 24, 31, 33] have long been of interest, because they enable one to carry out global conformational searches of proteins in real time, which in turn can facilitate ab initio protein structure predictions based solely on Anfinsen's thermodynamic hypothesis [1], according to which the native structure of a protein is a global minimum of its potential energy surface [32]. In the last few years, we developed a united-residue force field [20-22, 24], hereafter referred to as UNRES, in which a polypeptide chain is represented by a sequence of α-carbon (Cα) atoms linked by virtual bonds with attached united side chains (SC) and united peptide groups (p) located in the middle between the consecutive α-carbons (Figure 1). Only the united peptide groups and united side chains serve as interaction sites, the α-carbons serving to define the geometry.

The EDMC (electrostatically driven Monte Carlo) method has proven to be an effective computational tool for searching the potential energy hypersurface of polypeptide molecules consisting of up to 20 amino acid residues. Such a Monte Carlo search combined with gradient-based energy minimization of molecular conformations results in the need for 100 gigaflop or higher performance levels. The parallel EDMC algorithm has been designed to exploit currently available supercomputing technology. The implementation on the iPSC/2 described appears to represent an improvement over the original version for the IBM 3090. A performance analysis indicates that the attainable parallelism is limited by the underlying acceptance rate of search. It is demonstrated that a coarse-grained approach is suitable for architectures such as the CRAY-XMP, particularly if vectorization techniques can be exploited. Tests on the Intel iPSC/2-VX computer have shown, however, that even the easily vectorized parts of the computation may not overcome a large vector pipeline latency.>